ol

hilipaMax Planck Institute for Evolutionary Anthropology, DbAnthropology Ph.D. Program, City University of New YcNew York Consortium in Evolutionary Primatology, USd Paleoanthropology, Department of Early Prehistory andRmelinstr. 23, 72070 Tbingen, Germany

a r t i c l e i n f o

Article history:Received 24 December 2011Accepted 8 August 2012Available online 11 September 2012

features, variously aligning them with Homo erectus sensu lato, H.neanderthalensis and H. sapiens. The most complete faces include

breadth, and details of the nasal margin and palate (Stringer, 1974;Arsuaga et al., 1997; Rightmire, 1998, 2008; Trinkaus, 2003, 2006;Athreya, 2009). They also share several facial features with Nean-derthals, such as facial prognathism, broad nasal apertures, anda lack of concavity in the infraorbital region (Rak, 1986; Trinkaus,1987, 2003, 2006; Arsuaga et al., 1997; Harvati et al., 2010).

Striking differences between MPhs, Neanderthals and recentmodern humans occur in their facial size and robusticity. Inparticular, Bodo, Petralona and Kabwe are characterized by their

* Corresponding author.E-mail addresses: sarah.freidline@eva.mpg.de, sfreidline@gmail.com

(S.E. Freidline), gunz@eva.mpg.de (P. Gunz), katerina.harvati@ifu.uni-tuebingen.de(K. Harvati), hublin@eva.mpg.de (J.-J. Hublin).1 Present address: Max Planck Institute for Evolutionary Anthropology, Depart-

Contents lists available at

Journal of Hum

journal homepage: www.el

Journal of Human Evolution 63 (2012) 723e740ment of Human Evolution, Deutscher Platz 6, Leipzig 04103, Germany.differences between Neanderthals and Middle Pleistocene humans are due to scaling along a sharedallometric trajectory. These features include aspects of the frontal bone, browridge morphology, nasalaperture size and facial prognathism. Infraorbital surface topography and orientation of the midface inthe European Middle Pleistocene hominins is intermediate between the African Middle Pleistocene andNeanderthal condition. This could suggest that the European Middle Pleistocene hominins displayincipient Neanderthal features.

2012 Elsevier Ltd. All rights reserved.

Introduction

The Eurasian and African Middle Pleistocene hominins (MPh)are characterized by a combination of craniofacial morphological

Bodo and Kabwe (i.e., Broken Hill) from Africa, and Arago 21, Pet-ralona, and Sima de los Huesos 5 from Europe. MPh specimens arecharacterized bymassive facial bones, broad upper faces, projectingbrowridges, especially in the lateral region, large interorbitalKeywords:AllometryFaceOntogenyHomo heidelbergensisNeanderthalsSemilandmark geometric morphometrics0047-2484/$ e see front matter 2012 Elsevier Ltd.http://dx.doi.org/10.1016/j.jhevol.2012.08.002epartment of Human Evolution, Deutscher Platz 6, Leipzig 04103, Germanyork Graduate School, 365 Fifth Avenue, NY 10016, USAAQuaternary Ecology, Eberhard Karls Universitt Tbingen and Senckenberg Center for Human Evolution and Paleoecology,

a b s t r a c t

Neanderthals and modern humans exhibit distinct facial architectures. The patterning of facialmorphology of their predecessors, the Middle Pleistocene humans, is more mosaic showing a mix ofarchaic and modern morphologies. Signicant changes in facial size and robusticity occurred throughoutPleistocene human evolution, resulting in temporal trends in both facial reduction and enlargement.However, the allometric patterning in facial morphology in archaic humans is not well understood. Thisstudy explores temporal trends in facial morphology in order to gain a clearer understanding of thepolarity of features, and describes the allometric patterning of facial shape.The modern human sample comprises cross-sectional growth series of four morphologically distinct

human populations. The fossil sample covers specimens from the Middle Pleistocene to the UpperPaleolithic. We digitized landmarks and semilandmarks on surface and computed tomography scans andanalyzed the Procrustes shape coordinates. Principal component analyses were performed, andProcrustes distances were used to identify phenetic similarities between fossil hominins. In order toexplore the inuence of size on facial features, allometric trajectories were calculated for fossil andmodern human groups, and developmental simulations were performed.We show that facial features can be used to separate Pleistocene humans into temporal clusters. The

distinctly modern human pattern of facial morphology is already present around 170 ka. Species- andpopulation-specic facial features develop before two years of age, and several of the large-scale facialSarah E. Freidline a,b,c,d,*,1, P p Gunz a, Katerina Harvati b,c,d, Jean-Jacques Hublin aMiddle Pleistocene human facial morphand developmental contextAll rights reserved.ogy in an evolutionary

SciVerse ScienceDirect

an Evolution

sevier .com/locate/ jhevol

umamassive facial bones compared with Neanderthals and recentmodern humans (Rightmire, 1998), and both the MPhs and Nean-derthals have longer and more prognathic faces than recentmodern humans (Trinkaus, 2003). To what extent the facialdifferences among recent modern humans, MPhs, and Neander-thals are related to facial size is uncertain.

Allometry, as dened by Gould (1966), is the study of size and itsconsequences. Conventionally, it is used to investigate the rela-tionship between the total body size of the organism and its shape,anatomy, and physiology among other characteristics (Gould,1966). Julian Huxley (1932) introduced the concept of allometryinto studies of growth, evolution and function, and later S.J. Gouldsseminal work in both allometry (e.g., Gould, 1966, 1975) and het-erochrony (e.g., Gould, 1968, 1977), the dissociation of size, shapeand age (Gould, 1977; Alberch et al., 1979), set the framework forcomparative allometric ontogenetic investigations among livingand extinct organisms. Since Gould, a substantial body of workinvestigating heterochrony and growth allometry in human andnon-human primates has accumulated (e.g., Shea, 1981, 1983; Leighand Cheverud, 1991; Ravosa, 1991, 1998; Collard and OHiggins,2001; Penin et al., 2002; Leigh et al., 2003; Berge and Penin,2004; Strand Viarsdttir and Cobb, 2004; Mitteroecker et al.,2004, 2005; Leigh, 2006; Lieberman et al., 2007; Bruner andRipani, 2008; Gonzalez et al., 2011), however studies on fossilhominins are less common.

Rosas (1997, 2000) and Rosas and Bastir (2004) identiedseveral features on the Middle Pleistocene Sima de los Huesos (SH)mandibles associated with an increase in size, including the pres-ence of a retromolar space, the curvature and orientation of thesymphysis, and the position of the mental foramen with respect tothe dentition. Similarly, in their geometric morphometric study onmodern human and Neanderthal mandibular shape, Nicholson andHarvati (2006) found that the retromolar gap was related to anincrease in mandibular size. Size-correlated shape changes in laterhominins and recent modern humans have also been observed inthe relative breadth of the nasal aperture (Rosas and Bastir, 2002;Holton and Franciscus, 2008), nasoglabellar prole (Rosas andBastir, 2002), alveolar prognathism (Rosas and Bastir, 2002), andinfraorbital surface topography (Maddux and Franciscus, 2009). Inthe latter study, Maddux and Franciscus (2009) identied an allo-metric relationship between infraorbital size and surface topog-raphy within Middle to Late Pleistocene Homo and modernhumans. They found that individuals with large infraorbital regionstend to exhibit at, or inated, surface topographies (characteristicof Neanderthals), while individuals with small infraorbital areaspossess depressed surface topographies (like modern humans).They argued that depressed versus inated infraorbital shapes arenot dichotomous congurations, but fall along a continuous sizegradient. Together, these studies demonstrate that some Neander-thal apomorphies (e.g., retromolar space and infraorbital ination)are size-correlated.

However, a major limitation to several of the studies mentionedabove is the lack of ontogenetic data. Without a developmentalgrowth series, these studies can only demonstrate that there areallometric shape changes among adults (i.e., static allometry). Anunderstanding of the developmental basis of morphologicalfeatures is crucial to interpreting the taxonomic importance ofcharacters because features that develop earlier in ontogeny arethought to be less susceptible to epigenetic factors, or less likely torespond to exposure to environmental effects (Lieberman et al.,1996, 2002; Wood and Lieberman, 2001; but see; Roseman et al.,2010).

Due to a growing body of ontogenetic studies, it is becomingincreasingly clear that taxon-specic craniofacial morphology

S.E. Freidline et al. / Journal of H724among hominin species developed prenatally or very earlypostnatally (e.g., Ponce de Len and Zollikofer, 2001; Ackermannand Krovitz, 2002; Lieberman et al., 2002; Strand Viarsdttiret al., 2002; Williams et al., 2002; Krovitz, 2003; Bastir and Rosas,2004; Mitteroecker et al., 2004; McNulty et al., 2006; Gunz et al.,2010, 2012; Gunz, 2012). These morphometric studies largelysupport descriptive research identifying unique Neanderthalmorphology at an early age (e.g., Tillier, 1989, 2011; Rak et al., 1994;Akazawa et al., 1995). For example, at birth the ear labyrinth ofNeanderthals are different in size and shape when compared withmodern humans (Hublin et al., 1996; Ponce de Len and Zollikofer,1999), and by at least two years of age Neanderthal features on theoccipital, such as the suprainiac fossa and a bilaterally protrudingoccipital torus, have already developed (Hublin,1980). Additionally,studies have shown that at the time of birth, the face of a Nean-derthal is already larger than that of a modern human (Ponce deLen and Zollikofer, 2001; Ponce de Len et al., 2008; Zollikoferand Ponce de Len, 2010; Gunz et al., 2010, 2011, 2012). On theother hand, several Neanderthal facial features, such as midfacialprojection, maxillary ination, and double-arched browridges, arenot fully developed until later in childhood (Tillier, 1996).

Ponce de Len and Zollikofer (2001) showed that Neanderthalsand modern humans share a common pattern of cranial andmandibular shape change from an early age and onward. Thegeneral pattern of shape change comprised a projection anddownward elongation of the face and mandible combined witha contraction of the cranial vault. They concluded that the cranio-facial and mandibular differences between these two groupsprobably results from differential activity of growth elds early inontogeny (Ponce de Len and Zollikofer, 2001). Similarly, in theirgeometric morphometric study on facial development in great apesand Australopithecus africanus, Ackermann and Krovitz (2002)found that facial features arose very early in development fol-lowed by parallel postnatal developmental patterns (with thepossible exception of the gorillas). Additionally, they found thataspects of facial growthweremore similar between A. africanus andmodern humans, relative to the great apes. Therefore, their studysuggests that our early human ancestors were already demon-strating some human-like aspects of facial growth. The implicationsof these ndings are that one can interchange hominid postnatalgrowth trajectories without producing signicant differences in theend results. This has been further demonstrated by the study ofMcNulty et als. (2006) on the taxonomic afnities of the subadultTaung fossil specimen. To evaluate the adult morphology of theTaung child, McNulty et al. (2006) performed a series of develop-mental simulations. They grew Taung along various homininedevelopmental trajectories and compared its adult morphologywith adults of both Australopithecus and Paranthropus. WhileMcNulty et al. (2006) found the developmental patterns of extanthominine species to be statistically different, the results from theirdevelopmental simulations indicate that the postnatal develop-mental differences between hominines have little impact on theestimation of the adult morphology. McNulty et al. (2006)demonstrated that the adult morphology of Taung can be reliablyestimated even through the application of an incorrect develop-mental trajectory.

However, several geometric morphometric analyses on humansand primates have found that both early postnatal cranialmorphology and later postnatal growth contribute to furtherdifferentiate populations and species (Richtsmeier et al., 1993;OHiggins and Jones, 1998; OHiggins et al., 2001; StrandViarsdttir et al., 2002; Bastir and Rosas, 2004; Cobb andOHiggins, 2004; Strand Viarsdttir and Cobb, 2004; Bastir et al.,2007). For example, Bastir et al. (2007) argued that both pre- andpostnatal ontogenetic growth are important in establishing

n Evolution 63 (2012) 723e740morphological differences in mandibular shape between

umaNeanderthals and modern humans. Their results showed divergentontogenetic shape changes between Neanderthals and modernhumans and signicantly different allometric scaling patterns.

The aim of this study is to place the MPhs in both a geographi-cally and chronologically broad evolutionary and developmentalcontext to gain a clearer understanding of how archaic (i.e., MPhsand Neanderthals) and modern human facial features are affectedby facial size and how they change through time. More specically,the goals of this study are to: 1) explore temporal trends in facialmorphology in order to gain a clearer understanding of the polarityof facial features, 2) describe the shape changes associated withallometric scaling in archaic and modern human faces, and 3) testwhether differences in facial shape between archaic and modernhumans are attributable to the differential extension or truncationof common growth allometries (i.e., ontogenetic scaling), orwhether shape variation is due to divergent growth patterns.

Temporal trends in facial morphology are explored through themeans of a principal component analysis (PCA) of Procrustes shapevariables and Procrustes distance comparisons. The second andthird objectives are addressed by comparing the angle and orien-tation of Neanderthal and modern human ontogenetic allometrictrajectories in both shape and Procrustes form space, and byapplying developmental simulations (McNulty et al., 2006;Neubauer et al., 2010; Gunz et al., 2010, 2012; Gunz, 2012) to growmodern human andNeanderthal faces to the size of selectedMiddlePleistocene fossils. The predicted facial shapes are then visualizedand compared between archaic and modern human groups.

In this study, we recognize three categories of allometry:ontogenetic, static and evolutionary (Cock, 1966; Cheverud, 1982;Klingenberg, 1998). We use cross-sectional data on recent modernand Pleistocene humans to explore patterns in ontogenetic allom-etry, dened as the covariation of traits with size across differentage groups (i.e., ontogenetic stages) of a given species (Klingenberg,1998). Static allometry reects trait covariationwith size within thesame ontogenetic stage of a single species (Klingenberg, 1998). Thistype of allometry is observed among the adults in our sample.Lastly, evolutionary allometry arises from covariation of traits withsize and their phylogenetic changes and can be analyzed eitherwithin one ontogenetic stage or across stages (Klingenberg, 1998).In his inuential paper evaluating the relationship between thesethree types of allometry, Cheverud (1982), following work by Cock(1966), Gould (1966) and Shea (1981), demonstrated that patternsof static adult allometry cannot be assumed to reect ontogeneticprocesses (although see Klingenberg and Zimmermann, 1992;Klingenberg, 1998). This is mainly because adult data usuallyrepresent a very restricted (in both size and time) subset of theontogenetic range (Inouye and Shea, 1997).

Following Gould (1975), an allometric regression line can act asan ontogenetic criterion of subtraction such that the points alongthe line can be explained in terms of size-required changes. Devi-ations from the line are non-allometric shape changes and themorphological differences may indicate specic functional adap-tations (Gould, 1975). In this study, we used ontogenetic scaling asa criterion of subtraction because it is a powerful means for iden-tifying shared effects of size change between groups or species, aswell as derived dissociations of ancestral allometries (Gould, 1975;Shea, 1981, 1983; Inouye and Shea, 1997). Here we use a multivar-iate approach to allometric scaling that differs in certain method-ological aspects from the classical allometric studies. Departingfrom the conventional denition and applications of allometry, e.g.,we substitute facial size for body size because we are specicallyinterested in how differences in facial size inuence the expressionof facial features and overall facial shape. Surface semilandmarkgeometric morphometric techniques are used to quantify facial

S.E. Freidline et al. / Journal of Hfeatures that are otherwise difcult to capture, such as theinfraorbital surface topography, and developmental simulations areapplied to visualize the effects of allometry on the face.

Evolutionary models of Middle Pleistocene hominins

The two main competing phylogenetic interpretations of theMiddle Pleistocene human fossil record are: 1) most of the MPhscomprise a single, cross-continental taxon, spanning Africa, Europeand possibly Asia, or 2) the European and African MPhs, at leastthose fromMarine Isotope Stage 11 and onward, belong to separateclades. In the rst model, the MPhs are assigned to the taxon H.heidelbergensis and are considered to be the last common ancestorto both Neanderthals and modern humans. This view is supportedby the strong morphological and metric similarities between theEuropean and African specimens (e.g., Stringer, 1974, 1983; Arsuagaet al., 1997; Rightmire, 1998, 2007, 2008; Mounier et al., 2009;Harvati, 2009a).

In the second model, the European MPhs are seen as ancestralto Neanderthals and are classied as either H. neanderthalensis(Hublin,1998, 2009) or as the exclusivemembersofH.heidelbergensis,a chronospecies of the Neanderthal lineage (e.g., Arsuaga et al., 1997;Manzi, 2004; but seeWolpoff et al.,1994; Rosas et al., 2006; Tattersalland Schwartz, 2006; Bruer, 2008 for alternative interpretations ofthe fossil record). In this scenario, the AfricanMPhs are interpreted asbeing ancestral to H. sapiens and often classied as H. rhodesiensis(Hublin, 2009). As evidence for the Neanderthalization process,proponents of this model refer to a series of features on the skeletonthat foreshadow the Neanderthal condition and that occur uniquelyin the European MPhs (Dean et al., 1998; Hublin, 1998, 2009). Thesefeatures are found on the face, occiput, temporal bone and dentition.They include, but are not limited to, a convex and receding horizontalinfraorbital prole, anteriorly advanced and sagittally oriented face,wide occipital torus, incipient suprainiac fossa, bilaterally protrudingoccipital torus, strong juxtamastoid eminence (Dean et al., 1998;Hublin, 1998), and derived conditions expressed in different non-metrical traits of the dentition (Martinn-Torres et al., 2012).However, quantitative support for this model is limited. This may inpart be due to the difculty in quantifying these complex featuresespecially on the face (Harvati et al., 2010).

Materials and methods

Sample

This study includes a comprehensive sample of subadult andadultMiddle to Late Pleistocene fossil hominins (Table 1) and recentmodernhumans (Table 2). The fossil samplewas designed to includeall available Middle to Late Pleistocene fossils that preserve rela-tively complete faces (see section on Missing data reconstructionbelow). Table 1 lists the 26 fossils, their broad geographical loca-tion, chronology and repository.

The modern human sample (Table 2) comprises a cross-sectional growth series from four geographically diverse humanpopulations spanning three continents: Africa, North America andEurope. The individuals within each growth series range in agefrom two years to adulthood. The African sample consists of anarchaeological Khoisan population from South Africa. The NorthAmerican sample is divided into two groups: a combined NativeAmerican archaeological sample from Canyon del Muerto, Arizona,and Grand Gulch, Utah, and an archaeological population fromPoint Hope, Alaska. Lastly, the European sample consists ofa temporally more recent combined sample from Strasbourg(France) and Greifenberg (Austria). The modern human cranial datawere obtained from specimens housed in the American Museum of

n Evolution 63 (2012) 723e740 725Natural History (AMNH, New York), Iziko South African Museum

Table 1Fossil specimens used in the analysis, their abbreviation (Ab.), repository and chronology.a

Specimen Ab.b Repositoryc Chronology

Middle Pleistocene: AfricaBodo Bd NME ca. 600 ka (Clark et al., 1994)Kabwe Kb NHM 700e400 ka (Klein, 1994); late

Middle Pleistocene (Stringer, 2011)Middle Pleistocene: Europe

Arago 21d Ar UM 600e350 ka (Cook et al., 1982;Falgures et al., 2004)

Petralona Pt AUT 670eca. 250 ka (Harvati et al., 2009)Sima de los Huesos 5e Sm5 UCM ca. 530 ka (Bischoff et al., 2007)

Late MiddleeLate Pleistocene: AfricaA

MNPEHHH

PP

S.E. Freidline et al. / Journal of Human Evolution 63 (2012) 723e740726Jebel Irhoud 1 I1 MLate MiddleeLate Pleistocene: Europe

Gibraltar 1 Gb1 NHGuattari Gt MLa Chapelle-aux-Saints LCh MLa Ferrassie 1 LF1 MPech-de-lAz I Pech M

Late MiddleeLate Pleistocene: AsiaLiujiange Ljg IVQafzeh 6 Q6 UTQafzeh 9 Q9 UTShanidar 1e Sh1 IMShanidar 5e Sh5 IMSkhul 5 Sk5 PM(Cape Town), the University of Cape Town, and the Medical Facultyof Strasbourg.

Individual specimens in the human subadult skull collectionfrom the Medical Faculty of Strasbourg have known ages witha precision ranging from one month to one day (Rampont, 1994).Age estimates for all other subadult individuals were assessedaccording to dental eruption patterns following Ubelaker (1989).Each specimen in the modern human sample was classied bydevelopmental stage according to dental eruption sequence. Agegroup one is composed of subadults with a deciduous dentition,lacking the eruption (as dened by the exposure of cusps) of any

Table 2Recent modern human adult and subadult specimens used in the analysis.

Population/Geographic region AGa 1

Khoisan, South Africac,d,e 8Arizona (Canyon del Muerto), Utah (Grand Gulch), USAe 7Alaska (Point Hope), USAe 6Strasbourg, Francef and Greifenberg, Austriae 7

a The abbreviation AG represents Age Group. See text (Material and Methods e Sampb The specimen sex is denoted as M for males and F for females.c Iziko South African Museumd University of Cape Towne American Museum of Natural Historyf Medicine Faculty of Strasbourg

Teshik Tashe T-T MSU

Upper Paleolithic: EuropeCro-Magnon 1 e MHGrotte des Enfants 6e e MAPMladec 1 e NMPredmost 3e e DVM (

materOberkassel 1 e RLOberkassel 2 e RL

Upper Paleolithic: AsiaZhoukoudian 101d e Origin

Zhoukoudian 102d e Origin

a Subadult individuals are italicized.b Individual European and Asian Upper Paleolithic fossil material were not labeled inc The complete names of the repositories are available as a table in the SOM.d Casts from the Department of Human Evolution, Max Planck Institute for Evolutionae Casts from the Division of Anthropology, American Museum of Natural History (Newca. 160 ka (Smith et al., 2007)

71e50 to 35 ka (Klein, 1999)ca. 50 ka (Schwarcz et al., 1991)56e47 ka (Grn and Stringer, 1991)71e50 to 35 ka (Klein, 1999)51e41 ka (Soressi et al., 2007)

139e111 ka (Shen et al., 2002)135e100 ka (Grn et al., 2005)135e100 ka (Grn et al., 2005)ca. 50 ka (Trinkaus, 1983)ca. 50 ka (Trinkaus, 1983)135e100 ka (Grn et al., 2005)permanent teeth. Age group two is dened by the eruption of therst molar. Age group three is dened by the eruption of the secondmolar. Age group four is dened by the eruption of the third molar(i.e., complete set of permanent dentition). Table 2 lists the numberof subadults distributed among each age groupwithin eachmodernhuman population, and Table 3 lists the age estimates and group-ings for the subadult fossil specimens. Because of the difculty anduncertainty in identifying the sex of subadult individuals (Scheuerand Black, 2000), we made no attempt to do so, although studieshave shown that size and shape differences occur between malesand females throughout ontogeny and aremost pronounced during

AG 2 AG 3 AG 4 (adult) Total

8 4 38 (Mb: 14; F: 24) 586 4 52 (M: 25; F: 27) 6910 4 48 (M: 26; F: 22) 684 4 49 (M: 27; F: 22) 64

le) for denition of each age group.

ca. 70 ka (Movius, 1953); 57e24 ka(Vishnyatsky, 1999)

28e27 ka (Holt and Formicola, 2008)Gravettian (Henry-Gambier, 2001)ca. 31 ka (Holt and Formicola, 2008)

originalial missing)

Early Upper Paleolithic (Smith, 1982)

ca. 12 ka (Street, 2002)ca. 12 ka (Street, 2002)

al material missing ca. 13e33 (Chen et al., 1989;Brown, 1992; Hedges et al., 1992)

al material missing ca. 13e33 (Chen et al., 1989;Brown, 1992; Hedges et al., 1992)

the gures and therefore no abbreviation is provided for these specimens.

ry Anthropology (Leipzig, Germany).York).

each specimen and the Procrustes consensus conguration. Aftersliding, landmarks and semilandmarks were treated the same insubsequent statistical analyses. To convert the landmark andsemilandmark coordinates to shape variables, a generalizedProcrustes analysis (GPA) was performed. GPA removes the effectsof translation and rotation in the raw coordinate data and stan-dardizes each specimen to unit centroid size, the square root of thesum of squared distances from each landmark to the specimenscentroid (Dryden and Mardia, 1998). All data processing andstatistical analyses were performed in Mathematica (WolframResearch) and R (R Development Core Team, 2010).Missing data reconstruction As geometric morphometric methodsrequire all specimens to have the same number of homologouspoints, someminordata reconstructionwasnecessary for some fossilspecimens. First, bilateral symmetry was exploited by mirroring the

following Harvati (2001)Staphylion staZygomatic process root inferiorb zri The malar root origin

projected onto buccalalveolar surface

Zygomatic process root superiorb zrs The point where malarroot arises from themaxilla (often a pointof concavity betweenalveolare region andzygomaxillare) followingMcNulty (2003)

Zygomaxillareb zm

a All landmarks are dened following White et al. (2012). Denitions and refer-ences (if available) are provided for the less common landmarks

b Paired right and left landmarks

umaLandmark Editor software (Wiley et al., 2005). This templatemesh of surface semilandmarks was warped into the vicinity ofthe surface of every specimen using a thin-plate spline (TPS)interpolation according to the landmark and curve data. Thewarped points were then projected onto the surfaces by pickingand after adolescence (e.g., Strand Viarsdttir et al., 2002;Bulygina et al., 2006). Adult individuals were sexed according toHowells (1973) criteria, and when possible an equal number ofmales and females were included.

Measurement protocol

Computed tomography (CT) and surface scans CT or surface scansof the fossil and modern human crania were acquired. The CT scanswere made with either an industrial (BIR ACTIS 225/300) ormedical (Toshiba Aquilion) CT scanner, and the remaining speci-mens were surface scanned with either a Minolta Vivid 910 ora Breuckmann optoTOP-HE. The pixel size of the CT scans rangedfrom 0.24 to 0.49 mm and the slice thickness was between 0.25and 1.00 mm. The resolution of the surface scanners ranged fromw30 microns to w6 microns in the z plane. The differences inresolution among these scanners are much smaller than thedigitizing error and differences among the specimens. Forexample, although the adults and juveniles in the Europeansample were scanned using two different methods (adults:Minolta surface scanner, juveniles: medical CT scanner), theycluster together in a principal component analysis (PCA).

Three-dimensional surface models were extracted from eitherthe surface or CT scans and saved as .ply polygon model le format.Surface scans were processed using either Geomagic Studio orOptoCat (Breuckmann) software, depending on the scanner modelused. For the CT data, three-dimensional surfaces were extractedusing Avizo (Visualization Sciences Group Inc.). The landmarks andsemilandmarks were digitized on the surface models using Land-mark Editor (Wiley et al., 2005). If CT or surface scan data of theoriginal fossil material were not available, surface scans of highquality casts (see Table 1) from the Division of Anthropology of theAMNH (New York) or the Max Planck Institute for EvolutionaryAnthropology (Leipzig) were made.Landmark data Landmarks and semilandmarks, dening curvesand surfaces, were digitized by one observer (S.F.). 3D coordinatesof landmarks (Table 4) and semilandmarks along curves weredigitized on all specimens and a mesh of surface semilandmarkswas digitized on one template individual (see Fig. 1) using

Table 3Ontogenetic age and age group classication of the subadult fossil specimens used inthe analysis.a

Specimen Ontogenetic age (ca. yr.) Age group

Pech-de-lAz I 2 (Tillier, 1996) 1Teshik-Tash 9e11 (Tillier, 1989; Williams et al., 2002) 2Grotte des Enfants 6 13e15 (Henry-Gambier, 2001) 4

a See text (Material and Methods e Sample) for denition of age groupclassications.

S.E. Freidline et al. / Journal of Hthe closest vertices from the specimens .ply le following otherstudies (e.g., Gunz, 2005; Gunz and Harvati, 2007; Neubaueret al., 2009, 2010; Gunz et al., 2009a,b; Harvati et al., 2010;Stanseld and Gunz, 2011; Freidline et al., 2012). This protocolguarantees that every specimen has the same number of curvesemilandmarks and surface semilandmarks in approximatelycorresponding locations. A detailed description can be found inGunz et al. (2005, 2009b) and Mitteroecker and Gunz (2009).

The initially equidistant semilandmarks were slid alongtangents to the curves and tangent planes to the surfaces so as tominimize the bending energy of the TPS interpolation betweenTable 4Homologous landmarks used in the analysis.a

Landmark Abbreviation Denition

Alveolare idsAnterior nasal spine ans Thin projection of bone

on the midline at theinferior margin of thenasal aperture

Auriculareb auFrontomalare orbitaleb fmoFrontomalare temporaleb fmtFrontotemporaleb ftGlabella gJugaleb juMedial orbital marginb mm The point of maximal

concavity on the medialinferior corner of theorbital margin, wherethe orbital margin andthe frontal process ofthe maxilla meet

Nasion nNasospinale nsPorionb poRhinion rhiSphenopalatine sutureb ss Suture between palatine

and sphenoid bonesposterior and inferior tomaxillary tuberosity

n Evolution 63 (2012) 723e740 727surface of the better-preserved side along themidsagittalplaneusingGeomagic Studio and Avizo. The following specimens werecompleted by mirror-imaging: Bodo, Cro-Magnon 1, Gibraltar 1,Guattari, Kabwe, La Ferrassie 1, Oberkassel 1, and Shanidar 1.

Second, if missing data occurred bilaterally on a specimen oralong its midline, landmarks were estimated using geometricreconstruction via TPS interpolation (Gunz et al., 2009a). Thismethod has been applied in several other studies (e.g., Gunz, 2005,2012; Gunz andHarvati, 2007; Gunz et al., 2009a,b, 2010, 2011, 2012;Harvati et al., 2010; Bastir et al., 2011; Benazzi et al., 2011; Stanseldand Gunz, 2011; Freidline et al., 2012; Neubauer et al., 2012) to

umareconstruct fragmentary fossil material. Arago 21 required the mostextensive reconstruction. A detailed account of this reconstructionand the protocol can be found in Gunz et al. (2009b).

Statistical analyses

To explore the temporal variability in facial morphology amongPleistocene humans, a PCA was performed in shape space. A PCA

Figure 1. The landmark and semilandmark dataset. 671 landmarks and semiland-marks digitized on all specimens, red: homologous landmarks; blue: curve-semilandmarks; yellow: surface-semilandmarks. Homologous landmarks are abbre-viated. The full names are listed in Table 4. (For interpretation of the references tocolour in this gure legend, the reader is referred to the web version of this article.)

S.E. Freidline et al. / Journal of H728reduces the dimensionality of high dimensional shape space andprovides summaries of large-scale trends within the data(Bookstein, 1991; Rohlf, 1993). We used TPS warping of theProcrustes mean shape along the rst two principal components(PCs) to visualize the shape changes (see Gunz and Harvati, 2007;Mitteroecker and Gunz, 2009). Additionally, to identify whichindividuals are most phenetically similar to one another, nearestneighbors were calculated using inter-individual Procrustesdistances. To visualize the morphological similarities sharedbetween nearest neighbors, each of the Middle Pleistocene homi-nins was superimposed on its nearest neighbor.

In morphometrics, a distinction is made between shape andform. Shape refers to the geometric properties of an object that areindependent of its overall size, position and orientation, whereasthe form of an object includes both its shape and size (Mitteroeckerand Gunz, 2009). Because this study focuses on allometry, theinclusion of size in our statistical analyses is essential. To quantifyfacial size, we used the variable centroid size, which is computedduring the Procrustes superimposition. Following Mitteroeckeret al. (2004, 2005), we performed a PCA in Procrustes form space(also called size-shape space) on the Procrustes landmark andsemilandmark coordinates. Form space includes the geometric size(as the natural logarithm of centroid size) of each specimen(Mitteroecker et al., 2004) and is valuable because the relationshipbetween shape and size can be readily explored.

In this study, we compute the ontogenetic allometric trajectoriesas multivariate regressions of the Procrustes shape variables on thenatural logarithm of centroid size. As outlined byMitteroecker et al.(2004, 2005), properties of these trajectories (e.g., length, shape,divergence) can be compared between populations and species.Coincidental ontogenetic trajectories indicate a common pattern ofshared development between the groups under investigation. Weinterpret coincidental growth trajectories for the extant (recentmodern humans) and extinct groups (Neanderthals) as a sharedpattern of relative growth that also may be truncated or extended.Parallel trajectories reveal that the morphological divergencebetween groups has already occurred prior to the age of the youn-gest specimen, and that there is a common pattern of subsequentgrowth shared between groups. By contrast, divergent trajectoriesindicate that development further accentuates group differences.

Ontogenetic trajectories were plotted in shape and Procrustesform space for Neanderthals, each of the four modern humanpopulations, and the modern human mean. The modern humanmean trajectory was calculated by regressing Procrustes shapecoordinates on the natural logarithm of centroid size on the entiremodern human sample. Because only adult MPhs faces are includedin this study, static, rather than ontogenetic, allometric trajectorieswere calculated and plotted in Procrustes shape and form space forthese hominins.

To see if modern humans and Neanderthals share a commonontogenetic allometric trajectory during postnatal ontogeny, wecomputed the angle between the Neanderthal and the meanmodern human ontogenetic allometric vectors. We then testedwhether the angle was statistically different from zero usinga permutation test (McNulty et al., 2006; Gunz, 2012). We rstmean-centered the respective groups and then compared theactual angle with angles obtained from regressions computed for5000 permutations that randomly reassigned group membership,permuting the entire sample for each iteration. Likewise, the anglesbetween the ontogenetic allometric trajectories of the four modernhuman populations were calculated and tested to see if they weresignicantly different from zero. To assess the effects of the smallNeanderthal sample size on the computation of the ontogeneticallometric trajectory, we bootstrapped subsamples from the recentmodern human sample, simulating the composition of the Nean-derthal sample (i.e., one specimen from age group 1, one from agegroup 2 and six adults).

Developmental simulations and visualization techniques

The aim of the series of developmental simulations was toidentify the allometric shape changes that occur when scaling anadult modern human and Neanderthal to the size of a MPh. To doso, we calculated the mean adult modern shape and grew itfollowing the modern human mean ontogenetic allometric trajec-tory to the size of a) Kabwe, an African MPh that has a particularlyinated infraorbital surface topography for a MPh, and b) Bodo, anAfrican MPh that has the largest face in our sample. Shape changeswere visualized by TPS warping of the adult modern human meanshape to the modern human mean grown to the size of Bodo andKabwe. The same approach was applied to Neanderthals. We war-ped the surface of La Ferrassie 1 to the mean adult Neanderthalshape, and then grew it following the Neanderthal ontogeneticallometric trajectory to the size of Kabwe and Bodo.

Results

Principal component analyses and permutation tests

To explore temporal trends in facial morphology, a PCA wasperformed in Procrustes shape space of only the adult individuals(Fig. 2). The rst two principal components (PCs) represent 46.9% ofthe total shape variation, and neither of the PCs correlate with size.

n Evolution 63 (2012) 723e740There is a clear separation between the modern humans, including

umaS.E. Freidline et al. / Journal of Hthe early modern and Upper Paleolithic humans, and the archaichumans. Additionally, members of each temporal group clustertogether. However, there is some overlap between Middle Pleisto-cene humans and Neanderthals. Themain shape changes that occuralong PC 1 are in the curvature of the anterior portion of the frontalbone, browridge projection, width of the nasal aperture and facialprognathism. Middle Pleistocene specimens, such as Bodo andSima de los Huesos 5, and Neanderthals, such as La Chapelle-aux-Saints, cluster at the extreme positive end of PC 1. These fossilsshare a more receding anterior frontal bone, anteriorly projectingbrowridge, wide nasal aperture, and prognathic mid and lower facewhen compared with recent modern humans, which generally plotat the negative end of PC 1. While several shape changes are sharedbetween PC 1 and PC 2, such as frontal bone curvature, facialprognathism and infraorbital surface topography, PC 2 primarilydemonstrates changes in facial length and orientation of themaxillary body (i.e., maxillary body facies). Shanidar 5, Petralonaand Kabwe share a superiorlyeinferiorly long face and theirmaxillary body is oriented near vertically. On the opposite end of PC2, the modern humans express a shorter face and a maxillary bodythat is oriented down and slightly posteriorly.

Figure 2. PCA in shape space of adult fossil and modern humans. A convex hull is drawnnearest neighbor according to inter-individual Procrustes distances. The black arrows specifyfull names and abbreviations for the fossil specimens are listed in Table 1. The surface visun Evolution 63 (2012) 723e740 729Based on pairwise Procrustes distances (PD), Bodo is mostsimilar to Sima de los Huesos 5 (PD 0.068), Sima de los Huesos 5 ismost similar to Arago 21 (PD 0.06), Arago 21 is most similar to LaChapelle (PD 0.05), Petralona and Kabwe are most similar to oneanother (PD 0.056), and Shanidar 5 is most similar to Petralona(PD 0.056). To illustrate the resemblances between these speci-mens, we superimposed each MPh onto its nearest neighbor(Fig. 3aeo). Bodo and Sima de los Huesos 5 share similarly projec-ting lateral browridges, wide nasal apertures and interorbitalbreadths. However, Bodos glabellar region is more anteriorly pro-jecting and itsmid and lower facemorphology is especially differentfrom Sima de los Huesos 5 (Fig. 3aec). For example, Sima de losHuesos 5 has a more prognathic mid and lower face, a slightly moreinated infraorbital surface topography and a more parasagittallyoriented midface when compared with Bodo (Fig. 3c: Sima de losHuesos 5 is green and Bodo is bone color). Sima de los Huesos 5 andArago 21 (Fig. 3def) share a similar mid and lower face morphologyand diverging upper faces (i.e., browridges). Thus they sharea similar infraorbital surface topography, and they exhibit a similardegree of facial prognathism and parasagittal rotation. However,Arago 21 has a more projecting browridge, especially the middle

for the Upper Paleolithic modern humans. The arrows indicate each fossil specimensif an individual is most similar to another individual outside of its temporal group. Thealizations represent the mean shapes at the positive and negative ends of PC 1 and 2.

umaS.E. Freidline et al. / Journal of H730portion (Fig. 3f: Sima de los Huesos 5 is green). The similaritiesbetween Arago 21 and La Chapelle (Fig. 3gei) are in midfacialprognathism and infraorbital surface topography. La Chapelle hasamore anteriorly projecting browridge thanArago 21, except for thelateral portions, its midface is more parasagitally oriented, and itsnasal aperture is more anteriorly projecting (Fig. 3i: La Chapelle isred). Kabwe and Petralona (Fig. 3jel) primarily share similarities inthe lateralmidface,mainly in zygomatic bone projection. Kabwehasa more anteriorly projecting browridge, and a superiorlyeinferiorlylonger and more prognathic subnasal region. However, the

Figure 3. Procrustes superimposition of each of the Middle Pleistocene hominins on its necontained in the landmark and semilandmarks and everything in between these landmarkspecimens surface color (dark yellow is associated with the bone colored surfaces; dark greecolored landmarks are exposed this indicates overlapping (i.e., similar) morphology. Differesurface is visible. Nearest neighbors are indicated by the arrows in Fig. 2. The full names anreferences to colour in this gure legend, the reader is referred to the web version of thisn Evolution 63 (2012) 723e740infraorbital topography of Petralona is more inated than Kabwe(Fig. 3l: Kabwe is green). Lastly, Shanidar 5 and Petralona (Fig. 3meo) share a similar degree of mid and lower facial prognathism,infraorbital ination, and nasal aperturewidth. Themain differencebetween these two specimens is in the morphology of their brow-ridge. Petralona expresses a more anteriorly projecting browridgeand upper face (Fig. 3o: Shanidar 5 is red).

The results of the PCA in Procrustes shape space of the completesample (i.e., subadult modern humans and Neanderthals) areshown in Fig. 4. The rst three PCs were plotted and represent

arest neighbor in shape space. The Procrustes shape information for each specimen iss is interpolated. The landmarks and semilandmarks are color-coded according to thenwith the green surfaces; and dark red with the red surfaces). Thus, when two differentnt surface morphologies are present when only one landmark color and correspondingd abbreviations for the fossil specimens are listed in Table 1. (For interpretation of thearticle.)

Figure 4. PCA in shape space including subadult and adult fossil and modern humans. The rst three PCs are plotted. PC 1 represents 32.5% of total shape variance, PC 2 represents17.8%, and PC 3 represents 8.8%. The lines indicate the ontogenetic allometric trajectories for Neanderthals (in red), the mean modern human trajectory (in dark gray), and the

for td retraps to

S.E. Freidline et al. / Journal of Human Evolution 63 (2012) 723e740 73160.0% of total shape variance. PC 1 is correlated with size(r y 0.79). The solid lines represent the ontogenetic allometrictrajectories for each of the four modern human populations, theNeanderthal and mean modern human ontogenetic allometrictrajectories, and the MPh static allometric trajectory. The shadedregion represents the 95% single prediction interval of the modernhuman regressions of shape on the natural logarithm of centroidsize. It is evident that all Neanderthal and MPh adults fall outsidethis condence interval for allometric scaling. The thin lines

trajectory for each modern human population (in black). The static allometric trajectorylogarithm of centroid size. The specimen color-coding is the same as in Fig. 2. The shadeshape on the natural logarithm of centroid size. The thin gray lines represent the bootsthat simulate the Neanderthal sample composition. (For interpretation of the referencerepresent the bootstrapped estimates of the modern human allo-metric trajectory computed from subsamples that simulate theNeanderthal sample composition.

None of the Neanderthal or recent modern human ontogeneticallometric trajectories coincide. Because each ontogenetic

Figure 5. PCA in form space of subadults and adult fossil and modern humans. Form spacerepresents 79.0% of total form variance, PC 2 represents 5.0%, and PC 3 represents 3.3%. The lmodern human trajectory (in dark gray), and the trajectory for each modern human populatwere calculated by regressing form on the natural logarithm of centroid size. The specimeprediction interval of the modern human regressions of shape on the natural logarithm ofhuman allometric trajectory computed from subsamples that simulate the Neanderthal samhuman overgrown along the mean modern human ontogenetic allometric trajectory to thpretation of the references to colour in this gure legend, the reader is referred to the webtrajectory has a unique starting point, this plot shows that pop-ulation and species-specic facial morphology is present beforepermanent teeth erupt. As in Fig. 2, clusters of temporal groups arealso apparent when including the entire sample (i.e., all subadultsincluded) although the MPh and Neanderthals overlap. Apart fromQafzeh 9, the early and Upper Paleolithic modern humans fallwithin the range of recent human variation.

The variance explained by ontogenetic allometry is 26.0% forrecent modern humans and 59.7% for Neanderthals. The angle

he MPhs is in green. The trajectories were calculated by regressing shape on the naturalgion represents the 95% single prediction interval of the modern human regressions ofped estimates of the modern human allometric trajectory computed from subsamplescolour in this gure legend, the reader is referred to the web version of this article.)between the modern human average ontogenetic trajectory andthe Neanderthal ontogenetic trajectory in the subspace of the rstthree PCs is 28.0. A permutation test using all dimensions of shapespace, however, reveals that the slopes of the Neanderthal andmean modern human (MH) ontogenetic trajectories are not

includes the log centroid size for each individual. The rst three PCs are plotted. PC 1ines indicate the ontogenetic allometric trajectories for Neanderthals (in red), the meanion (in black). The static allometric trajectory for the MPhs is in green. The trajectoriesn color-coding is the same as in Fig. 2. The shaded region represents the 95% singlecentroid size. The thin gray lines represent the bootstrapped estimates of the modernple composition. The two circles demonstrate where the predicted shapes of a moderne size of Kabwe and Bodo plot in the PCA (see text for more information). (For inter-version of this article.)

signicantly different from zero (p < 0.31). Therefore, we cannotreject the null hypothesis that the mean modern human andNeanderthal ontogenetic allometric trajectories are parallel.

To further explore the effects of size on facial morphology, weperformed a PCA in Procrustes form space (Fig. 5). The rst threePCs represent 85.0% of total variance. The Neanderthal and modernhuman mean ontogenetic trajectories do not coincide. The anglebetween the two trajectories in the subspace of the rst three PCs is8.8. However, a permutation test cannot reject the null hypothesisthat these trajectories are parallel (p< 0.35). As expected in a form-

facial shape and its ontogenetic allometric trajectory. Thesedevelopmental simulations are depicted in Fig. 7def. Table 6 liststhe facial features affected by allometry in both recent modernhumans and Neanderthals. Among the recent modern humans inour sample, the Point Hope group has the largest faces accordingcentroid size. Therefore, to verify that this population was notdriving our results, we removed them and recalculated the meanmodern human developmental allometric trajectory. The sameallometric shape changes occurred in modern humans when thePoint Hope sample was not included. The following facial shape

ma

mer

< 0

0.00.0

r) an

S.E. Freidline et al. / Journal of Human Evolution 63 (2012) 723e740732space analysis, PC 1 is highly correlated with centroid size(r 0.99). The variance explained by ontogenetic allometry inrecent modern humans is 80.1% and in Neanderthals it is 92.4%. Asin shape space, the shaded region in Fig. 5 represents the 95% singleprediction interval of themodern human regressions of form on thenatural logarithm of centroid size, and, as in shape space, allNeanderthal and MPh adults fall outside this modern humancondence interval for allometric scaling. The thin lines representthe bootstrapped estimates of the modern human allometrictrajectory computed from subsamples that simulate the Neander-thal sample composition.

The Neanderthal infant Pech de LAz and the youngest recentMH subadult individuals cluster at the left end of the plot (indi-cating smaller size) and the individuals with larger faces, e.g., Bodo,Petralona, and Kabwe fall at the opposite end. The MPhs fall at theend of the Neanderthal ontogenetic allometric trajectory and theorientation of the Neanderthal ontogenetic and MPh static trajec-tories appear similar. The position of the larger MPhs, Kabwe,Petralona and Bodo suggests that allometric scaling alonga common trajectory explains some differences in facialmorphology between them and Neanderthals. The early MH andUpper Paleolithic humans cluster with the adult recent MH.

Table 5 lists the angles between the ontogenetic allometrictrajectories for each of the four modern human populations. Theangles were computed in the subspace of the rst three PCs tomake them easier to interpret. The permutation tests werecomputed using all dimensions of shape space and Procrustes formspace, respectively. The results of the permutation test indicate thatin both shape and form space, the trajectories of each of themodernhuman populations is signicantly different from zero (i.e., notparallel). The only exceptions are the angles between the PointHope and Khoisan populations (shape space: angle 4.3, p< 0.23;form space: angle 18.5, p < 0.07).

Developmental simulations

To visualize the allometric shape changes that occur in the faceof recent modern humans, the modern human mean shape wascalculated and grown to the size of Bodo, the largest specimen inour sample, following the modern human mean ontogenetic allo-metric trajectory. Fig. 6def illustrates the allometric shape changesthat occur in recent modern humans. Developmental simulationswere also performed on Neanderthals using the Neanderthal mean

Table 5Angles between the ontogenetic allometric trajectories of the four recent modern hu

Africa North A

Africa 0 23.7 (pNorth Americab 6.3 (p < 0.006) 0Europe 8.4 (p < 0.0005) 5.3 (p